Open the Sync image/audio upload widget so a user can choose a ChatGPT file or upload a supported local image/audio file, then stage it as a durable Sync assetId. Use this by default when the user wants to upload or choose a local image/audio file and has not attached the file to ChatGPT yet. For image-to-speech requests, call this with requestedMediaType: "image" and tell the user to enter the exact requested text in the widget Script field. This widget cannot accept video or MP4 files. Never use, recommend, or describe this tool for local video requests, and never use requestedMediaType: "video". For local video or MP4, ask the user to attach the video to the ChatGPT composer, then call upload-media or create-lipsync with file params.

Upload a user-provided image, video, or audio file to Sync asset storage and return a durable assetId. Use this when the user uploaded media in chat and a later Sync tool call should reference it by assetId. The file field must be the uploaded ChatGPT file object. For public URLs, use assets_create or pass the URL directly to create-lipsync. This tool only stores the media; it does not create a lipsync generation. After it returns, pass the assetId to create-lipsync as imageAssetId, videoAssetId, or audioAssetId.

Create a lipsync video from audio + EITHER a video or a still image (an image drives sync-3 image-to-video). Defaults to sync-3 unless the user explicitly requests another model. Generations are attached to an existing project with the requested projectName, or to "ChatGPT generations" by default; if no matching project exists, it is created first. For "make this image/video say X" requests, pass script with a voiceId from voices_get-voices; do not call tts_create first. Pass URLs whenever you have them — set audioUrl to any public audio URL, and videoUrl/imageUrl to a hosted media URL. If media was uploaded to Sync first, pass audioAssetId, videoAssetId, or imageAssetId. For files the user uploaded in chat, prefer calling upload-media first and pass the returned assetId here; direct audio/video/image file params are supported only when a host invokes this tool with file params directly. If the user wants to choose a local image/audio file that is not attached yet, use open-upload-widget first. Provide exactly one visual input (video or image) and exactly one driver input (audio or script). Returns a generation id — call generate_get-generation once with wait: true and timeout: 55, then read outputUrl. Copy signed outputUrl values exactly from the tool result; never reconstruct or shorten them. For advanced options (segments, speaker selection), use generate_create-generation.

List assets (video/audio/image) in your organization. Use to find an existing assetId.

Register a media URL as a reusable asset (e.g. the url from assets_create-upload-url, or any public URL). Returns an asset id to use as input[].assetId in a generation, or as a voice-clone sample.

Upload step 1: get a presigned URL for a local file. PUT the raw bytes to the returned uploadUrl (same Content-Type), then register it with assets_create using the returned url. Use this when your media is not already at a public URL. Max 5GB.

Get one asset by id (details + URL).

Update an asset's name or visibility by id.

Delete an asset by id.

Create a lipsync video. Provide a video input (or an image for sync-3) and an audio input, each by url or assetId. Returns a generation id — call generate_get-generation once with wait: true and timeout: 55, then read outputUrl. To choose which face to sync in a multi-person video, pass options.active_speaker_detection (auto_detect, or coordinates + frame_number).

Get a generation by id. In ChatGPT, pass wait: true and timeout: 55 for generation results instead of making repeated manual polls. When status is COMPLETED, copy the exact structuredContent.outputUrl string verbatim; signed URLs must not be reconstructed, shortened, or edited.

Estimate the credit cost of a generation before creating it. Takes the same body as generate_create-generation.

List recent generations for your organization. Use to find generation ids or check status.

List the lipsync models available to you (e.g. lipsync-2, lipsync-2-pro, sync-3, react-1). Use a returned model id as the model in generate_create-generation.

Estimate the credit cost of a generation before creating it. Returns estimated credits.

Get a generation by id (organization-scoped). Returns status and, when COMPLETED, outputUrl.

Delete a generation by id. Only terminal generations (COMPLETED/FAILED/REJECTED) can be deleted — deleting one that is still processing returns 409.

List the projects in your organization.

Create a project to group related generations and assets. Pass the returned id as projectId on generate_create-generation or assets_create so they show up together in Studio.

Attach an asset to a project

Detach an asset from a project

Get a project by id.

Update a project's name, description, visibility, or mode.

Delete a project. Generations and assets attached to it are not deleted — they just stop being grouped under it.

Synthesize standalone speech audio from text. Use only when the user specifically asks for an audio file or voice preview. Do not use this for "make this image/video say X" lipsync requests — use create-lipsync with script and voiceId directly.

List available voices — premade ElevenLabs voices plus voices your org has cloned. For "make this image/video say X", use a returned voice id as voiceId in create-lipsync with script directly. Use tts_create only when the user explicitly asks for standalone audio.

Clone a custom voice from an audio or video sample — pass a Sync-hosted url or an assetId (upload local files via assets_create-upload-url first), plus provider (elevenlabs) and a name. Returns a voice id for tts_create.

Get a voice by ID

Delete a cloned voice by id, freeing a clone slot.